Standardized contrast-enhanced CT-based radiomics for non-invasive prediction of TTF-1 status in lung adenocarcinoma: A cross-validated single-center study.
Authors
Affiliations (3)
Affiliations (3)
- Institute of Radiology, Neuroradiology and Interventional Therapy, Klinikum Bayreuth, Medical Campus Upper Franconia, Friedrich Alexander University, Erlangen, Germany.
- Department of Nuclear Medicine, Klinikum Bayreuth, Bayreuth, Germany.
- Institute of Pathology, Klinikum Bayreuth, Bayreuth, Germany.
Abstract
Thyroid transcription factor-1 (TTF-1) is a clinically relevant immunohistochemical marker in lung adenocarcinoma with diagnostic, prognostic, and therapeutic implications. As tissue sampling is invasive and often limited by small biopsy specimens, non-invasive imaging biomarkers are of increasing interest. This study aimed to develop and internally validate a standardized contrast-enhanced CT-based radiomics workflow for prediction of TTF-1 expression, with emphasis on reproducibility and methodological robustness. This retrospective single-center study included 164 treatment-naive patients with histologically confirmed lung adenocarcinoma and available TTF-1 immunohistochemistry. TTF-1 expression was assessed using routine immunohistochemical staining on biopsy or resection specimens, with binary classification based on nuclear staining. Primary tumors were semi-automatically segmented on contrast-enhanced CT using IBSI-compliant preprocessing. A total of 107 radiomic features were extracted. Interobserver reproducibility was evaluated in a subset of 55 tumors using intraclass correlation coefficients. Logistic regression, random forest, and XGBoost models were trained using five-fold stratified cross-validation. Performance was evaluated using AUC. Among 164 patients, 115 tumors (70.1%) were TTF-1 positive. XGBoost achieved the highest performance (mean AUC = 0.83 ± 0.04), significantly outperforming other models (p < 0.01). Radiomic features showed good interobserver reproducibility, particularly for shape and first-order features. SHAP analysis identified sphericity, entropy, and mean CT attenuation as key predictors. A standardized CT-based radiomics approach shows promising performance for non-invasive prediction of TTF-1 status in lung adenocarcinoma. Given the retrospective single-center design and lack of external validation, further multicenter studies are required.